Re-add Pearson/Spearman correlation based filtering of features

[dhanak]

Strongly correlated features can mess up the screening, because if A and B appear randomly half the time, then the single core feature they both represent appears to be half as important as it really is.

There was some code related to this filtering earlier in the repostiory, but it got cleaned up.